ToolsDeveloper visualized costs of agent models, highlighting Luna's cheapnessA developer created their own visualizer of the costs of the larger models used in today's agents, with data sourced from Artificial Analysis.1 min read
ToolsOpenAI improves GPT‑6 prompt caching to boost hit rates and lower costsOpenAI updated prompt caching for the GPT‑6 family to increase cache hit rates by default, offer discounts for reused shared prefixes within a 30‑minute window, and reduce latency for persistent agents.3 min read
Toolsllm 0.36 released with new OpenAI GPT‑6 models and conversation support controlsThe llm library version 0.36 adds recognition for two new OpenAI models, gpt‑6‑sol (GPT‑6 Sol) and gpt‑6‑luna (GPT‑6 Luna), and introduces a plugin flag to mark single‑turn models that do not support conversations.2 min read
ToolsNVIDIA Confidential Computing Enables Private AI Inference with Minimal Performance LossNVIDIA evaluated confidential AI inference on Blackwell-based B200 GPUs and found that Confidential Computing (CC) preserves most production throughput when frameworks adapt to secure execution.5 min read
ToolsTopology-aware scheduling with NVIDIA Topograph for improved GPU utilizationNVIDIA Topograph is an open-source toolkit that discovers and normalizes GPU- and fabric-topology information from cloud APIs and on‑premises fabrics, publishing scheduler-ready outputs for Kubernetes, Slurm and Slinky.6 min read
Toolsllm-anthropic 0.29 released: support for the Claude Opus 5.5 modelllm-anthropic version 0.29 has been released, extending the package's support to the Anthropic Claude Opus 5.5 language model.1 min read
ToolsOpus 5.5 faster and cheaper to port than Fable 5.1Both the Opus 5.5 and Fable 5.1 models ported HAProxy from C to Rust; both passed most tests, but Opus 5.5 completed the process in 9.5 hours versus Fable 5.1's 12 hours, and resulted in 51% lower…1 min read
Toolsllm-typesafe 0.1a0 plugin adds support for TypeSafe AI's Jev modelA new plugin, llm-typesafe 0.1a0, extends the LLM tool to support TypeSafe AI's Jev model.2 min read
ToolsGPT‑6 Astra halved Parallel’s research time and code costsParallel tested GPT‑6 Astra on a multi‑state labor‑market research task and found the model produced the same quality outputs in roughly half the time and with about a 50% reduction in code cost compared to prior models.2 min read
ToolsSpeeding ROS 2 Nodes with rosidl::Buffer and a CUDA Buffer BackendNVIDIA and upstream ROS 2 introduced a CUDA-backed rosidl::Buffer and a corresponding buffer backend in ROS 2 Lyrical that let nodes exchange GPU-resident message payloads without CPU copies or serialization when runtime conditions allow.5 min read
ToolsAstroForge to Fly Transformer-Based Autonomy on Future SpacecraftAstroForge, an asteroid-mining startup founded in 2022 and backed by $56 million in venture funding, is developing an onboard autonomous control stack called Solo to operate spacecraft with minimal ground support.3 min read
ToolsTransformers adds GGUF support for efficient local inference on Apple SiliconHugging Face's transformers now supports loading and running GGUF quantized checkpoints efficiently on Apple Silicon by reusing ggml Metal kernels.6 min read
